Blind Multiuser Detector Based On Ica Under Impulse Noise

CDMA (Code Division Multiple Access) has become the main access of the third generation mobile communication system (3G). However, when the number of users or the signal power increases, MAI is very serious, this greatly limits the CDMA system performance. Multi-user detection utilizes the internal structure information of user signature to improve reception performance of the system, in order to eliminate mutual interference among multiple users. In 1979, KS Schncider first proposed the idea of MUD (multi-user detector), a variety of multi-user detections become a research hotspot in recent years. But the noise channel based on the previous MUD methods is the Gaussian distribution. In some transient and great spikes occasions, the distributions of these signals are much thicker than the tail of Gaussian distribution, and with infinite variance. At this time, the methods based on original second-order statistics no longer worked, it needs new restrictions from the parameters of variance model approach.This paper introduces the ICA (Independent Component Analysis) method in multi-user detection, simulates the current traditional blind MUD algorithm based on second order statistics and analyzes advantages and disadvantages of these algorithms. After that, Gaussian noise model is instead of non-Gaussian noise model which is based on FLOS (Fractional lower Order Statistic) processing method. This paper separately filters the non-Gaussian noise in signals with two methods:CMA (Constant Modulus Algorithm) blind MUD based on FLOS and Infomax blind MUD based on Fractional Order Pre-whitening. Because the second order statistics does not exist in Alpha stable distributions, the original algorithm needs to be improved. The paper improves original in CMA blind MUD algorithm based on FLOS and first processes signal with the method of Fractional Order Pre-whitening in Infomax blind MUD algorithm based on Fractional Order Pre-whitening. The computer simulations indicate that the performance of the two methods is superior to traditional blind MUD algorithm based on second order statistics.